Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

New Mac Pro's Nehalem's "hyper Threading" Is A Major Issue With Pro Tools! Read This

I am a long time ProTools user and I am not shocked over the news.

Digidesign has always been slow to update their software. It always takes them at least an entire year to catch up, and when they do, Apple releases something new.

If you are depending on your rig to make money, then you would know to not always be an early adopter of the latest and greatest (due to the high risk of possibilty for incompatibilities).
 
I'm not shocked at all.
Digi is always behind the curve like that, (not bashing, it's just true) and I'd imagine that there may be at least some good reasons for that.
'specially since all the major DAW's seem to be taking some time catching up.

I've been saying this in the audio forums for awhile, that it'd take dev's some time to get things humming on this new architecture.
Which is why I've been (and continue to) saying that the '08 is the best choice for audio stuff if somebody is wanting it to work now.
 
mm.
I've read more than one post now, reading bad news about freezes, not (yet) supported hardware for the programs I use the most...
I'm cancelling my order.
I'll just try to find an '08. I need to work NOW without probs.
 
dazz said:
mm.
I've read more than one post now, reading bad news about freezes, not (yet) supported hardware for the programs I use the most...
I'm cancelling my order.
I'll just try to find an '08. I need to work NOW without probs.
Click to expand...

Not hatin' on the new ones, but that's almost certainly a good idea.
 
rockinrocker said:
I'm not shocked at all.
Digi is always behind the curve like that, (not bashing, it's just true) and I'd imagine that there may be at least some good reasons for that.
'specially since all the major DAW's seem to be taking some time catching up.
Click to expand...


Just to make it clear,that is what their business is.
They feed the extremely high priced pros with their prog/setup systems and it is natural that they want to make it sure that most of the things go by the book..
Pro tools is de facto system in most of the recording studios and they just cant loose the battle.


Too bad apple is not performing at the same level with FCP and Logic.
 
I see that as of yesterday Pro Tools HD is qualified with the Nehalem Mac Pros. Hopefully that means they will roll something out for us poor LE users. This makes my MP buying decision even harder now.
 
Being a Pro Tools user you should know better then to do any updates/upgrades until Digidesign has released an update for it. Computer, OS what ever.
 
synth3tik said:
Being a Pro Tools user you should know better then to do any updates/upgrades until Digidesign has released an update for it. Computer, OS what ever.
Click to expand...

Good point, thanks for the reality check, I'm getting the 2008 8 core. I've still got the headache of Avid Media Composer breaking our Firestore a/v sync...I should have known better:eek:
 
Sorry, don't have time to read the whole thread... but I just wanted to say that we're seeing similar problems with our in-house massively parallel code.

On my new Nehalem Mac Pro... OSX will happily schedule multiple parallel processes to the same physical core... and leave physical cores completely inactive.

For instance if I fire up an 8 process parallel run... I have seen the scheduler leave _3_ physical cores empty... loading up 3 cores with two processes each! (I actually have screenshots of this...).

Linux figured out many years ago that this is not ideal... and modified their scheduler to keep this from happening (and it works well).

We're in contact with our Apple representative... trying to get this worked out. Right now it's not clear how long it's going to take Apple to fix this...

BTW - A short term fix for any apps that have severe problems with this (like the Pro apps mentioned in the original post) is to turn off Hyperthreading using the CPUPalette app (under /Library » Application Support » HWPrefs). Obviously this isn't ideal... but it will keep bad stuff from happening.

Friedmud
 
synth3tik said:
Being a Pro Tools user you should know better then to do any updates/upgrades until Digidesign has released an update for it. Computer, OS what ever.
Click to expand...

Hell yeah..... God how long did it take before they even supported Leopard, god help us all when Snow Leopard comes out :D
 
friedmud said:
Sorry, don't have time to read the whole thread... but I just wanted to say that we're seeing similar problems with our in-house massively parallel code.

On my new Nehalem Mac Pro... OSX will happily schedule multiple parallel processes to the same physical core... and leave physical cores completely inactive.

For instance if I fire up an 8 process parallel run... I have seen the scheduler leave _3_ physical cores empty... loading up 3 cores with two processes each! (I actually have screenshots of this...).

Linux figured out many years ago that this is not ideal... and modified their scheduler to keep this from happening (and it works well).

We're in contact with our Apple representative... trying to get this worked out. Right now it's not clear how long it's going to take Apple to fix this...

BTW - A short term fix for any apps that have severe problems with this (like the Pro apps mentioned in the original post) is to turn off Hyperthreading using the CPUPalette app (under /Library » Application Support » HWPrefs). Obviously this isn't ideal... but it will keep bad stuff from happening.

Friedmud
Click to expand...

Not too surprising; Intel's hack of pretending that hyperthreading results in more cpu cores is dumb, especially for OS's that treat all "cores" as equal. Hopefully we won't have to wait for SL to fix this.
 
cmaier said:
Not too surprising; Intel's hack of pretending that hyperthreading results in more cpu cores is dumb, especially for OS's that treat all "cores" as equal. Hopefully we won't have to wait for SL to fix this.
Click to expand...

Actually... it's not that simple anymore. Because of the cache architecture in Nehalem's vs P4's.... a lot of the benchmarks we're doing really are showing that in a lot of cases a Hyperthreaded core is "almost" equal to a real core! Our theory is that the big shared L3 is reducing the penalty for context switches that occurred with Hyperthreading on P4's.

Like I say... this isn't always the case (as is hinted at by this thread)... but we're definitely not seeing the Hyperthread performance hit from the P4 days...

Friedmud
 
nanofrog said:
YES. :) I know it can be a lot of reading, but generally, there's a decent chance the information is in the thread. ;) It was in this case. :D

Skimmed too quickly perhaps. ;) I would imagine most of us have been guilty of it on occasion. :p
Click to expand...

Indeed. I find people often overlook things I've said in the post direction above their own... in my case, I poured hot water over my laptop and it's now running in safemode (640x480), so I have little screen real-estate.. I didn't even realize this was more than a 1-page thread. -shrugs-

Anyway, I wasn't asking because I needed to know, I was asking because it would obviously help some people if it's possible.
 
Digi sucks, I'm a Logic user but at a lot of studios we use PT and I also sell Digidesign gear. Never seen any DAW with so many issues concerning hardware setups, software operating system, etc etc.

In one instance there was a bluetooth mouse conflicting with pro tools :eek: In any given studio, if a Pro Tools setup works.. we don't change it one bit! No updates, no upgrades - nada!

And don;t get me started with their 'Mbox 2 PRO' - nothing PRO about it. Cheap build quality - 1 out of 3 units have the front knobs broken (they keep rotating) - and I am talking brand new, shipped from the factory.

More like poop tools :p
 
Firefly2002 said:
Indeed. I find people often overlook things I've said in the post direction above their own... in my case, I poured hot water over my laptop and it's now running in safemode (640x480), so I have little screen real-estate.. I didn't even realize this was more than a 1-page thread. -shrugs-

Anyway, I wasn't asking because I needed to know, I was asking because it would obviously help some people if it's possible.
Click to expand...

My bad ;)
Headache + assuming everyone reads at least the last page in a thread = grumpy ;)
 
friedmud said:
Actually... it's not that simple anymore. Because of the cache architecture in Nehalem's vs P4's.... a lot of the benchmarks we're doing really are showing that in a lot of cases a Hyperthreaded core is "almost" equal to a real core! Our theory is that the big shared L3 is reducing the penalty for context switches that occurred with Hyperthreading on P4's.

Like I say... this isn't always the case (as is hinted at by this thread)... but we're definitely not seeing the Hyperthread performance hit from the P4 days...

Friedmud
Click to expand...

If a core is saturated, sending anything to the second virtual core that shares the core can only slow things down, is my point. Three pipelines is three pipelines. Intel didn't double-up on the ALU's to allow two things to happen at once. As you said, the second core relies on context switches. If you have a parallelizable app where each thread has no dependencies, and if there are few cache misses and page faults, you're better off not having the virtual cores at all.
 
I do support for a couple studios in my area, so this is good info.

Digidesign is always late with support. They will get it worked out eventually. Most of the guys I know in the field don't upgrade very often anyway.
 
cmaier said:
If a core is saturated, sending anything to the second virtual core that shares the core can only slow things down, is my point. Three pipelines is three pipelines. Intel didn't double-up on the ALU's to allow two things to happen at once. As you said, the second core relies on context switches. If you have a parallelizable app where each thread has no dependencies, and if there are few cache misses and page faults, you're better off not having the virtual cores at all.
Click to expand...


Absolutely agree... I was just pointing out that in our testing it's not as bad as we thought it would be when two processes get scheduled to the same core. We were working off of our past experience with hyperthreading... And were surprised when the speed hit wasn't similar to the P4.

But yeah... The scheduler needs to get a freaking clue and at least not do stupid junk...
 
friedmud said:
On my new Nehalem Mac Pro... OSX will happily schedule multiple parallel processes to the same physical core... and leave physical cores completely inactive.

For instance if I fire up an 8 process parallel run... I have seen the scheduler leave _3_ physical cores empty... loading up 3 cores with two processes each! (I actually have screenshots of this...).
Click to expand...

MacRumorUser said:
.....god help us all when Snow Leopard comes out :D
Click to expand...

I think Macrumoruser hit the point.

Kind-of late to the conversation, but I think this is what Snow Leopard's BIG THING that everybody hypes about will be all about. Just getting the OS to interface with programs with all of the damn processors in a logical fashion. Hyperthread in a manner that the rest of us would see as the right thing to do.

If Apple does not manage to fix this problem in SL and destroy compatibilities with some smart, current programs (as with Friedmud's case), I will be ten kinds of :mad: for everybody else (lol, as if I could upgrade anyway on the Power platform).
 
nateDEEZY said:
People who actually use hardware don't stack plug-ins.
Click to expand...

Nonsense, not every effect has a hardware counterpart or solution. I work with some of the top mixing engineers in the industry, stacking plugins is common. I know 2 guys with the 09 MacPros and they don't seem to be having any issues. These are HD systems not LE...
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back